Artificial Intelligence and Robotics Commons^™

Plant Disease Detection Through Convolutional Neural Networks: A Survey Of Existing Literature, Best Practices, And Implementation, Kevin Label

Master's Theses

In the United States alone, common diseases spread among plants account for billions of dollars lost in crop yield each year. This issue is exacerbated in countries with less infrastructure to defend against crop epidemics, and can lead to famine and forced migration. Farmers can seek the help of plant pathology experts to defend against diseases and detect crop irregularities early on. However, access to experts can be difficult, and even those trained in the field may miss symptoms before it is too late. To assist in early disease detection, a number of papers have been released on the potential …

Toward Deep Supervised Anomaly Detection: Reinforcement Learning From Partially Labeled Anomaly Data, Guansong Pang, Anton Van Den Hengel, Chunhua Shen, Longbing Cao

Research Collection School Of Computing and Information Systems

We consider the problem of anomaly detection with a small set of partially labeled anomaly examples and a large-scale unlabeled dataset. This is a common scenario in many important applications. Existing related methods either exclusively fit the limited anomaly examples that typically do not span the entire set of anomalies, or proceed with unsupervised learning from the unlabeled data. We propose here instead a deep reinforcement learning-based approach that enables an end-to-end optimization of the detection of both labeled and unlabeled anomalies. This approach learns the known abnormality by automatically interacting with an anomalybiased simulation environment, while continuously extending the …

Fine-Grained Analysis Of Structured Output Prediction, Waleed Mustafa, Yunwen Lei, Antoine Ledent, Marius And Kloft

Research Collection School Of Computing and Information Systems

In machine learning we often encounter structured output prediction problems (SOPPs), i.e. problems where the output space admits a rich internal structure. Application domains where SOPPs naturally occur include natural language processing, speech recognition, and computer vision. Typical SOPPs have an extremely large label set, which grows exponentially as a function of the size of the output. Existing generalization analysis implies generalization bounds with at least a square-root dependency on the cardinality d of the label set, which can be vacuous in practice. In this paper, we significantly improve the state of the art by developing novel high-probability bounds with …

Optimization And Machine Learning Methods For Solving Combinatorial Problems In Urban Transportation, Aigerim Bogyrbayeva

USF Tampa Graduate Theses and Dissertations

This dissertation investigates three applications of emerging technologies for urban trans- portation. In the first chapter, we design a new market for fractional ownership of au- tonomous vehicles (AVs), in which an AV is co-leased by a group of individuals. We present a practical iterative auction based on the combinatorial clock auction to match the interested customers together and determine their payments. In designing such an auction, we con- sider continuous-time items (time slots) which are defined by bidders, and naturally exploit driverless mobility of AVs to form co-leasing groups. To relieve the computational burdens of both bidders and the …

Machine Learning In Stock Price Prediction Using Long Short-Term Memory Networks And Gradient Boosted Decision Trees, Carl Samuel Cederborg

Honors Projects

Quantitative analysis has been a staple of the financial world and investing for many years. Recently, machine learning has been applied to this field with varying levels of success. In this paper, two different methods of machine learning (ML) are applied to predicting stock prices. The first utilizes deep learning and Long Short-Term Memory networks (LSTMs), and the second uses ensemble learning in the form of gradient tree boosting. Using closing price as the training data and Root Mean Squared Error (RMSE) as the error metric, experimental results suggest the gradient boosting approach is more viable.

Honors Symposium: ML is …

Analog Spiking Neural Network Implementing Spike Timing-Dependent Plasticity On 65 Nm Cmos, Luke Vincent

Graduate Theses and Dissertations

Machine learning is a rapidly accelerating tool and technology used for countless applications in the modern world. There are many digital algorithms to deploy a machine learning program, but the most advanced and well-known algorithm is the artificial neural network (ANN). While ANNs demonstrate impressive reinforcement learning behaviors, they require large power consumption to operate. Therefore, an analog spiking neural network (SNN) implementing spike timing-dependent plasticity is proposed, developed, and tested to demonstrate equivalent learning abilities with fractional power consumption compared to its digital adversary.

Using Deep Learning To Automate The Diagnosis Of Skin Melanoma, Akhil Reddy Alasandagutti

Honors Theses

Machine learning and image processing techniques have been widely implemented in the field of medicine to help accurately diagnose a multitude of medical conditions. The automated diagnosis of skin melanoma is one such instance. However, a majority of the successful machine learning models that have been implemented in the past have used deep learning approaches where only raw image data has been utilized to train machine learning models, such as neural networks. While they have been quite effective at predicting the condition of these lesions, they lack key information about the images, such as clinical data, and features that medical …

Low-Power And Reconfigurable Asynchronous Asic Design Implementing Recurrent Neural Networks, Spencer Nelson

Graduate Theses and Dissertations

Artificial intelligence (AI) has experienced a tremendous surge in recent years, resulting in high demand for a wide array of implementations of algorithms in the field. With the rise of Internet-of-Things devices, the need for artificial intelligence algorithms implemented in hardware with tight design restrictions has become even more prevalent. In terms of low power and area, ASIC implementations have the best case. However, these implementations suffer from high non-recurring engineering costs, long time-to-market, and a complete lack of flexibility, which significantly hurts their appeal in an environment where time-to-market is so critical. The time-to-market gap can be shortened through …

Estimating Heading From Optic Flow With Neural Networks, Natalie T. Maus

Honors Theses

Humans have a remarkable ability to estimate their direction of self-motion, or heading, based on visual input stimulus (optic flow). Machines, on the other hand, have a difficult time with this task, especially when flow is introduced that is inconsistent with the motion of the observer. For example, when moving objects enter the field of view, their motion provides inconsistent flow data which often disrupts heading estimates of current heading estimation models. We investigate the ability of neural networks to estimate heading from optic flow data and the limitations of these models when different variations of inconsistent flow are introduced. …

Improving Space Efficiency Of Deep Neural Networks, Aliakbar Panahi

Theses and Dissertations

Language models employ a very large number of trainable parameters. Despite being highly overparameterized, these networks often achieve good out-of-sample test performance on the original task and easily fine-tune to related tasks. Recent observations involving, for example, intrinsic dimension of the objective landscape and the lottery ticket hypothesis, indicate that often training actively involves only a small fraction of the parameter space. Thus, a question remains how large a parameter space needs to be in the first place — the evidence from recent work on model compression, parameter sharing, factorized representations, and knowledge distillation increasingly shows that models can be …

A Deep Learning U-Net For Detecting And Segmenting Liver Tumors, Vidhya Cardozo

Theses and Dissertations

Visualization of liver tumors on simulation CT scans is challenging even with contrast-enhancement, due to the sensitivity of the contrast enhancement to the timing of the CT acquisition. Image registration to magnetic resonance imaging (MRI) can be helpful for delineation, but differences in patient position, liver shape and volume, and the lack of anatomical landmarks between the two image sets makes the task difficult. This study develops a U-Net based neural network for automated liver and tumor segmentation for purposes of radiotherapy treatment planning. Non-contrast simulation based abdominal CT axial scans of 52 patients with primary liver tumors were utilized. …

Nonlinear Dimensionality Reduction For The Thermodynamics Of Small Clusters Of Particles, Aditya Dendukuri

Graduate Theses and Dissertations

This work employs tools and methods from computer science to study clusters comprising a small number N of interacting particles, which are of interest in science, engineering, and nanotechnology. Specifically, the thermodynamics of such clusters is studied using techniques from spectral graph theory (SGT) and machine learning (ML). SGT is used to define the structure of the clusters and ML is used on ensembles of cluster configurations to detect state variables that can be used to model the thermodynamic properties of the system. While the most fundamental description of a cluster is in 3N dimensions, i.e., the Cartesian coordinates of …

Predictive Modeling Of Asynchronous Event Sequence Data, Jin Shang

LSU Doctoral Dissertations

Large volumes of temporal event data, such as online check-ins and electronic records of hospital admissions, are becoming increasingly available in a wide variety of applications including healthcare analytics, smart cities, and social network analysis. Those temporal events are often asynchronous, interdependent, and exhibiting self-exciting properties. For example, in the patient's diagnosis events, the elevated risk exists for a patient that has been recently at risk. Machine learning that leverages event sequence data can improve the prediction accuracy of future events and provide valuable services. For example, in e-commerce and network traffic diagnosis, the analysis of user activities can be …

Finding Critical And Gradient-Flat Points Of Deep Neural Network Loss Functions, Charles Gearhart Frye '09

Doctoral Dissertations

Despite the fact that the loss functions of deep neural networks are highly non-convex, gradient-based optimization algorithms converge to approximately the same performance from many random initial points. This makes neural networks easy to train, which, combined with their high representational capacity and implicit and explicit regularization strategies, leads to machine-learned algorithms of high quality with reasonable computational cost in a wide variety of domains.

One thread of work has focused on explaining this phenomenon by numerically characterizing the local curvature at critical points of the loss function, where gradients are zero. Such studies have reported that the loss functions …

Automatic Flood Detection In Sentinei-2 Images Using Deep Convolutional Neural Networks, Pallavi Jain, Bianca Schoen-Phelan, Robert J. Ross

Conference papers

The early and accurate detection of floods from satellite imagery can aid rescue planning and assessment of geophysical damage. Automatic identification of water from satellite images has historically relied on hand-crafted functions, but these often do not provide the accuracy and robustness needed for accurate and early flood detection. To try to overcome these limitations we investigate a tiered methodology combining water index like features with a deep convolutional neural network based solution to flood identification against the MediaEval 2019 flood dataset. Our method builds on existing deep neural network methods, and in particular the VGG16 network. Specifically, we explored …

Orthogonal Recurrent Neural Networks And Batch Normalization In Deep Neural Networks, Kyle Eric Helfrich

Theses and Dissertations--Mathematics

Despite the recent success of various machine learning techniques, there are still numerous obstacles that must be overcome. One obstacle is known as the vanishing/exploding gradient problem. This problem refers to gradients that either become zero or unbounded. This is a well known problem that commonly occurs in Recurrent Neural Networks (RNNs). In this work we describe how this problem can be mitigated, establish three different architectures that are designed to avoid this issue, and derive update schemes for each architecture. Another portion of this work focuses on the often used technique of batch normalization. Although found to be successful …

Multimodal Fusion Strategies For Outcome Prediction In Stroke, Esra Zihni, John D. Kelleher, Vince I. Madai, Ahmed Khalil, Ivana Galinovic, Jochen Fiebach, Michelle Livne, Dietmar Frey

Conference papers

Data driven methods are increasingly being adopted in the medical domain for clinical predictive modeling. Prediction of stroke outcome using machine learning could provide a decision support system for physicians to assist them in patient-oriented diagnosis and treatment. While patient-specific clinical parameters play an important role in outcome prediction, a multimodal fusion approach that integrates neuroimaging with clinical data has the potential to improve accuracy. This paper addresses two research questions: (a) does multimodal fusion aid in the prediction of stroke outcome, and (b) what fusion strategy is more suitable for the task at hand. The baselines for our experimental …

Finding A Viable Neural Network Architecture For Use With Upper Limb Prosthetics, Maxwell Lavin

Master of Science in Computer Science Theses

This paper attempts to answer the question of if it’s possible to produce a simple, quick, and accurate neural network for the use in upper-limb prosthetics. Through the implementation of convolutional and artificial neural networks and feature extraction on electromyographic data different possible architectures are examined with regards to processing time, complexity, and accuracy. It is found that the most accurate architecture is a multi-entry categorical cross entropy convolutional neural network with 100% accuracy. The issue is that it is also the slowest method requiring 9 minutes to run. The next best method found was a single-entry binary cross entropy …

Adversarial Attack On Neural Machine Translation System, Abijith K. P.

Master’s Dissertations

Nowadays Deep Neural Network based solutions are deployed to solve numerous tasks. Thus, it has become absolutely important to study the robustness of these systems. Machine Translation is one of the popular applications of Deep Neural Networks. This thesis studies the robustness of Neural Machine Translation systems by generating adversarial examples with the objective to fool the model. Whenever there is a change in the source, i.e. when a word in the input sentence is replaced by an unrelated word, the translation system is supposed to reflect the changes while doing translation. These unwanted invariance learned by the model is …

Do It Like A Syntactician: Using Binary Gramaticality Judgements To Train Sentence Encoders And Assess Their Sensitivity To Syntactic Structure, Pablo Gonzalez Martinez

Dissertations, Theses, and Capstone Projects

The binary nature of grammaticality judgments and their use to access the structure of syntax are a staple of modern linguistics. However, computational models of natural language rarely make use of grammaticality in their training or application. Furthermore, developments in modern neural NLP have produced a myriad of methods that push the baselines in many complex tasks, but those methods are typically not evaluated from a linguistic perspective. In this dissertation I use grammaticality judgements with artificially generated ungrammatical sentences to assess the performance of several neural encoders and propose them as a suitable training target to make models learn …

Labeling Paths With Convolutional Neural Networks, Sean Wallace, Kyle Wuerch

Computer Engineering

With the increasing development of autonomous vehicles, being able to detect driveable paths in arbitrary environments has become a prevalent problem in multiple industries. This project explores a technique which utilizes a discretized output map that is used to color an image based on the confidence that each block is a driveable path. This was done using a generalized convolutional neural network that was trained on a set of 3000 images taken from the perspective of a robot along with matching masks marking which portion of the image was a driveable path. The techniques used allowed for a labeling accuracy …

Dish: Democracy In State Houses, Nicholas A. Russo

Master's Theses

In our current political climate, state level legislators have become increasingly impor- tant. Due to cuts in funding and growing focus at the national level, public oversight for these legislators has drastically decreased. This makes it difficult for citizens and activists to understand the relationships and commonalities between legislators. This thesis provides three contributions to address this issue. First, we created a data set containing over 1200 features focused on a legislator’s activity on bills. Second, we created embeddings that represented a legislator’s level of activity and engagement for a given bill using a custom model called Democracy2Vec. Third, we …

Explainable Neural Attention Recommender Systems, Omer Tal

Theses and Dissertations (Comprehensive)

Recommender systems, predictive models that provide lists of personalized suggestions, have become increasingly popular in many web-based businesses. By presenting potential items that may interest a user, these systems are able to better monetize and improve users’ satisfaction. In recent years, the most successful approaches rely on capturing what best define users and items in the form of latent vectors, a numeric representation that assumes all instances can be described by their respective affiliation towards a set of hidden features. However, recommendation methods based on latent features still face some realworld limitations. The data sparsity problem originates from the unprecedented …

Predicting Post-Procedural Complications Using Neural Networks On Mimic-Iii Data, Namratha Mohan

LSU Master's Theses

The primary focus of this paper is the creation of a Machine Learning based algorithm for the analysis of large health based data sets. Our input was extracted from MIMIC-III, a large Health Record database of more than 40,000 patients. The main question was to predict if a patient will have complications during certain specified procedures performed in the hospital. These events are denoted by the icd9 code 996 in the individuals' health record. The output of our predictive model is a binary variable which outputs the value 1 if the patient is diagnosed with the specific complication or 0 …

Deep Machine Learning For Mechanical Performance And Failure Prediction, Elijah Reber, Nickolas D. Winovich, Guang Lin

The Summer Undergraduate Research Fellowship (SURF) Symposium

Deep learning has provided opportunities for advancement in many fields. One such opportunity is being able to accurately predict real world events. Ensuring proper motor function and being able to predict energy output is a valuable asset for owners of wind turbines. In this paper, we look at how effective a deep neural network is at predicting the failure or energy output of a wind turbine. A data set was obtained that contained sensor data from 17 wind turbines over 13 months, measuring numerous variables, such as spindle speed and blade position and whether or not the wind turbine experienced …

Deep Neural Network Architectures For Modulation Classification Using Principal Component Analysis, Sharan Ramjee, Shengtai Ju, Diyu Yang, Aly El Gamal

The Summer Undergraduate Research Fellowship (SURF) Symposium

In this work, we investigate the application of Principal Component Analysis to the task of wireless signal modulation recognition using deep neural network architectures. Sampling signals at the Nyquist rate, which is often very high, requires a large amount of energy and space to collect and store the samples. Moreover, the time taken to train neural networks for the task of modulation classification is large due to the large number of samples. These problems can be drastically reduced using Principal Component Analysis, which is a technique that allows us to reduce the dimensionality or number of features of the samples …

Extractive Text Summarization With Deep Learning, Garrett G. Chan

Computer Engineering

This project explores extractive text summarization using the capabilities of Deep Learning. The goal of this project is to create an application with a neural network to take in text as its input, and create a summary that is a shorter, condensed version of the input text. This has been implemented in Python by configuring and training a neural network that takes in a vector of features that are extracted from the text using various Natural Language Processing libraries. The implementation demonstrates that we can train simple deep neural networks to successfully summarize text.

Walknet: A Deep Learning Approach To Improving Sidewalk Quality And Accessibility, Andrew Abbott, Alex Deshowitz, Dennis Murray, Eric C. Larson

SMU Data Science Review

This paper proposes a framework for optimizing allocation of infrastructure spending on sidewalk improvement and allowing planners to focus their budgets on the areas in the most need. In this research, we identify curb ramps from Google Street View images using traditional machine learning and deep learning methods. Our convolutional neural network approach achieved an 83% accuracy and high level of precision when classifying curb cuts. We found that as the model received more data, the accuracy increased, which with the continued collection of crowdsourced labeling of curb cuts will increase the model’s classification power. We further investigated a model …

Target Detection Using Convolutional Neural Networks, Robert P. Loibl

Theses and Dissertations

This research explores the use of Convolutional Neural Networks (CNNs) to classify targets of interest within satellite imagery. Methods were specifically devised for the classification of airports within Landsat-8 scenes. A novel automated dataset generation technique was developed to create labeled datasets from satellite imagery using only coordinate metadata. Using this approach a very large dataset of over 132,000 labeled images was created without human input. This dataset was used to evaluate the effects of color and resolution on airport classification accuracy. Two experiments were run with the first experiment classifying large airports with 96.8% accuracy, and the second classifying …

Deep Neural Networks For Multi-Label Text Classification: Application To Coding Electronic Medical Records, Anthony Rios

Theses and Dissertations--Computer Science

Coding Electronic Medical Records (EMRs) with diagnosis and procedure codes is an essential task for billing, secondary data analyses, and monitoring health trends. Both speed and accuracy of coding are critical. While coding errors could lead to more patient-side financial burden and misinterpretation of a patient’s well-being, timely coding is also needed to avoid backlogs and additional costs for the healthcare facility. Therefore, it is necessary to develop automated diagnosis and procedure code recommendation methods that can be used by professional medical coders.

The main difficulty with developing automated EMR coding methods is the nature of the label space. The …